SIAP meeting : Industrial Image Analysis Wednesday, 6th December 2000 2-5pm RSS, 12 Errol Street, London EC1Y 8LX ______________________________________________________________________ TALK 1: Some problems in the study of magnetic materials D. M. Titterington (University of Glasgow) Abstract: The talk will cover a few aspects of the behaviour of thin-film magnetic materials under the influence of an external field. The materials themselves are used in such applications as modern information storage devices and in recording heads in hard disk drives. A short introduction to the elements of the underlying physics will be attempted, followed by descriptions of stochastic models for dynamic magnetisation reversal and of an empirical application of simultaneous autoregression models to analyse images of so-called magnetisation ripple. The work is the consequence of collaboration with colleagues in statistics and physics. ---------------------------- TALK 2: Image Analysis within Unilever Research: Quantifying Product Microstructure Geert M.P. van Kempen (Unilever Research) Abstract: It is widely believed that the microstructure of the composites typical of Unilever products plays a key role in the manifestation of the physical observables and behaviour of the materials and ultimately the consumer experience with the product. This belief has lead to a need for superior microstructural understanding in Unilever?s research activities, with the expectation that this will lead to superior product innovation and design. This presentation will give an overview of our efforts to quantify product microstructures and to discriminate between microstructures of products classes that differ on the basis of processing, ingredients or bulk physical properties. These efforts can be divided in three parts: the translation of microstructural attributes in parameters that can be measured by image analysis, classification of the microstructures on the basis of the measured parameters, and the identification and quantification of the found differences. In the first part of the presentation, the detection of minute differences in the microstructure of milk gels will be used to illustrate the concepts of the microstructure quantification toolbox we are currently developing. In the second part of the presentation, the focus will be on the development of advanced image analysis procedures to measure microstructure parameters. The measurement of the anisotropy and pore-sizes of monoglyceride gels demonstrate multi-scale image analysis. Both the monitoring of the dynamics of phase separation as well as the measurement of the gas cell size distribution during the proofing of dough will illustrate our efforts in time-resolved image analysis. Finally, the segmentation of the various phases inside a detergent powder particle shows the potential of multi-spectral analysis for microstructure quantification. In this particular case, we have developed a combined classifier, based on the EM algorithm, that uses information from both the spectral as well as the spatial domain to obtain improved segmentation results. The image analysis developments discussed in this presentation are developed in close collaboration with the Pattern Recognition Group of the Delft University of Technology. ---------------------------- TALK 3: Industrial Image Analysis: On the interaction between research and application Bjarne Kjaer Ersboell (Technical University of Denmark) Abstract: An element of research is of paramount importance for an industrial image analysis project to be interesting for a university department. The Section for Image Analysis has been involved in numerous industrial image analysis projects during the last twenty years. All are characterised by the research being application driven rather than method driven. In the process of solving the problem new theoretical aspects which need to be solved inevitably turn up. We find this interaction between theory and practice extremely inspiring. In the talk a small representative selection of industrial image analysis projects will be shown with special emphasis on the research performed. Finally, we will focus on a project on sorting of wood-slabs using image analysis. Here a combination of traditional statistical techniques together with some novel theory on feature selection was necessary in order to solve the problem. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%